“Large language models such as ChatGPT are emerging as powerful tools that not only make workers more productive but also increase the rate of innovation, laying the foundation for a significant acceleration in economic growth,” wrote economists Martin Neal Bailey, Erik Brynjolfsson, and Anton Korinek in “Machines of mind: The case for an AI-powered productivity boom,” a Brookings Institutions report. “However, official statistics will only partially capture the boost in productivity because the output of knowledge workers is difficult to measure. The rapid advances can have great benefits but may also lead to significant risks, so it is crucial to ensure that we steer progress in a direction that benefits all of society.”
The authors noted that generative AI systems are machines of the mind, increasingly capable of generating coherent and contextually appropriate text, images, videos, and audio, in a wide range of information-based tasks. Millions of knowledge workers, — from doctors and lawyers to managers and sales people, — will experience ground-breaking improvements in their productivity within a few years.
Let me summarize some of the key points in the Brookings report.
The power of productivity growth.
The article starts out by reminding us that productivity growth, the amount of output created per hour of work, is the primary determinant of long term prosperity. The key driver of changes in productivity growth is total factor productivity (TFP), a measure of the impact of technology on productivity.
From the end of WWII until the early 1970s, TFP grew at around 2.2% per year, a period that ushered an era of prosperity for most Americans. TFP growth then slowed dramatically to around 0.6% per year until the mid 1990, when thanks to the internet boom, it rebounded to 1.5% per year for the next decade. But, since 2005, TFP growth has been minimal, just around 0.4% per year. TFP growth has been particular anemic at 0.16% in the service sector of the economy, which accounts for almost 80% of US employment.
Early estimates of AI’s productivity effect.
“Generative AI has broad applications that will impact a wide range of workers, occupations, and activities,” wrote the authors. “Because information and knowledge work dominates the US economy, these machines of the mind will dramatically boost overall productivity.”
The report references “GPTs Are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models,” a recent research paper that analyzed the implications of LLMs on the US labor market. Based on the alignments of occupations with both GPT-4 capabilities and human expertise, the study found that LLMs could affect 10% or more of the tasks of 80% of the US workforce, and 50% of the tasks of close to 20% of workers. LLMs will likely impact all wage levels, with higher-income jobs facing greater exposure.
GenAI will improve productivity growth in two major ways. First, by increasing the efficiency and level of output of cognitive workers. And second, and ultimately more important by accelerating innovation and future productivity growth.
“Cognitive workers not only produce current output but also invent new things, engage in discoveries, and generate the technological progress that boosts future productivity. This includes R&D — what scientists do — and perhaps more importantly, the process of rolling out new innovations into production activities throughout the economy — what managers do. If cognitive workers are more efficient, they will accelerate technological progress and thereby boost the rate of productivity growth — in perpetuity.” Furthermore, because productivity growth compounds over the years, even a small increase in productivity growth would leave the economy significantly larger after several years.
Barriers and drivers of adoption.
“The Productivity J-Curve,” a working paper co-authored by Brynjolfsson, explained that general purpose transformative technologies, — e.g., the steam engine, electricity, computers, the internet, — have great potential from the outset, but history shows that they often failed to significantly improve productivity in their early years. Realizing their potential requires major complementary investments including business process redesign, innovative new products, applications and business models, the re-skilling of the workforce, and a fundamental rethinking of the very nature of production. As a result, there’s generally been a significant time lag, — sometimes decades, — between the marketplace appearance of a transformative technology and its widespread deployment and productivity growth across companies, industries, and economies.
General purpose physical technologies, — like electricity, for example, — required major physical infrastructures, including power lines, new kinds of motors, redesigned factories, the development of new electric household products, and so on. But, this is not the case with digital technologies like generative AI. These are being rolled out via software, with much of the required digital infrastructure already in place, including the internet, cloud computing, software-as-a-service, and the ability to integrate with existing applications. This lowers the time, efforts, expertise, and expense needed to embrace and deploy new digital technologies like genAI.
Problems of measurement – silent productivity growth
Another potential barrier is the difficulty of measuring the economic impact of generative AI on the productivity of cognitive jobs, mostly because of the difficulty of measuring labor productivity in the service sector in general. Gross Domestic Product (GDP) is a good measure of economic progress for an industrial economy dominated by the production of tangible physical products which are relatively easy to account for. But services now constitute a large portion of GDP and jobs around the world, — close to 80% in the US and other advanced economies. Due to their intangible nature, services are a kind of hard-to-measure dark matter. Perhaps the one definition everyone can agree to is one attributed to The Economist: a service is “anything sold in trade that cannot be dropped on your foot.”
“Statisticians who compile GDP and productivity statistics sometimes resort to valuing the output of cognitive activity simply by assuming it is proportional to the quantity of labor input being used to produce it, which of course eliminates any scope for productivity growth,” said the Brookings report. “For example, generative AI enables economists to write more thought pieces and provide deeper analyses of the economy than before, yet this output would not directly show up in GDP statistics.” Better and deeper economy analyses may well improve the productivity of business leaders and policymakers, but such positive effects of genAI would not be directly captured in official GDP or productivity statistics. “The same holds true for many other cognitive workers throughout the economy. This may give rise to significant under-measurement or silent productivity growth.”
Productivity growth, labor markets, and income distribution.
Over the past three decades, high skill jobs requiring expert problem solving and complex communication skills significantly expanded, with the earnings of the college educated workers needed to fill such jobs rising steadily. Conversely, opportunities and wages declined for less educated middle skill blue-collar manufacturing jobs and white-collar administrative jobs whose careers had been upended by automation.
But, the current wave of cognitive automation marks a change from these earlier waves of automation. Could generative AI end up reducing the growing job inequality of the past several decades? As a recent article in the NY Times noted: “The jobs most exposed to automation now are office jobs, those that require more cognitive skills, creativity and high levels of education. The workers affected are likelier to be highly paid, and slightly likelier to be women, a variety of research has found.”
“Large language models and other forms of generative AI are still at an early stage, making it difficult to predict with great confidence the exact productivity effects they will have,” said the Brookings report in conclusion. “Yet as we have argued, we expect that generative AI will have tremendous positive productivity effects, both by increasing the level of productivity and accelerating future productivity growth.”
“AI-powered productivity growth will also create challenges. There may be a need for updating social programs and tax policy to soften the welfare costs of labor market disruptions and ensure that the benefits of AI give rise to shared prosperity rather than concentration of wealth.”
“Economists and other social scientists will need to accelerate their work on AI’s impacts to keep up with our colleagues in AI research who are rapidly advancing the technologies. If we do that, we are optimistic our society can harness the productivity benefits and growth acceleration delivered by artificial intelligence to substantially advance human welfare in the coming years.”
Comments